There's plenty of examples of mathematicians like de Bourcia, who proved the Bieberbach conjecture in 1984 but also produced several manuscripts not accepted by the community, including a purported proof of the Riemann hypothesis.
The current failure mode of AI-generated math proofs is pages and pages of output that likely is slop but also might be correct.
In a sense this isn't a new problem.
Users are optimistic that future AI could filter math ideas to separate valuable proofs from unverified human slop.
Perhaps a future AI will be able to go through all the human slop and separate the wheat from the chaff. I suspect it will find that mathematicians were mostly correct to ignore it but did every now and then overlook great ideas too.
Wikipedia's entry on de Branges is somewhat hilarious.
"Mathematicians remain skeptical, and neither proof has been subjected to a serious analysis. [...] [I]t remains to be seen whether de Branges has managed to circumvent their objections."
Wikipedia's entry on de Bourcia is somewhat hilarious.
"Mathematicians remain skeptical, and neither proof has been subjected to a serious analysis. [...] [I]t remains to be seen whether de Branges has managed to circumvent their objections."
https://en.wikipedia.org/wiki/Louis_de_Branges_de_Bourcia
Perhaps a future AI will be able to go through all the human slop and separate the wheat from the chaff. I suspect it will find that mathematicians were mostly correct to ignore it but did every now and then overlook great ideas too.
Wikipedia's entry on de Bourcia is somewhat hilarious.
"Mathematicians remain skeptical, and neither proof has been subjected to a serious analysis. [...] [I]t remains to be seen whether de Branges has managed to circumvent their objections."
https://en.wikipedia.org/wiki/Louis_de_Branges_de_Bourcia
There's plenty of examples of mathematicians like de Bourcia, who proved the Bieberbach conjecture in 1984 but also produced several manuscripts not accepted by the community, including a purported proof of the Riemann hypothesis.
The current failure mode of AI-generated math proofs is pages and pages of output that likely is slop but also might be correct.
In a sense this isn't a new problem.